Recommend
 
 Thumb up
 Hide
9 Posts

Race for the Galaxy: Rebel vs Imperium» Forums » General

Subject: AI Question rss

Your Tags: Add tags
Popular Tags: [View All]
J R

New Jersey
msg tools
mbmbmbmb
So I recently downloaded the AI. Fantastic implementation, by the way. I've been playing 2 player advanced, and am winning about 90% of the time. I'd like to be able to offer some input on the AI, but it would help if I could get a fuller log after the game that showed the cards each player had in their hand at any given time. It would also be helpful if the discarded homeworld and the initial goals were revealed as well. Is this possible?

Here is a log of a recent game. I don't know what my opponent had, but slowing to a crawl with the explore calls seemed irrational from my perspective:

Red starts with Alpha Centauri.
Blue starts with New Sparta.
Red chooses Consume-Trade/Produce.
Blue chooses Explore +1,+1/Settle.
Red draws 2 and keeps 1.
Blue draws 3 and keeps 2.
Red pays 1 for Trading Outpost.
Blue conquers New Survivalists.
Red trades good from Alpha Centauri for 5.
Red receives 5 cards for Consume phase.
Red produces on Alpha Centauri.
Blue produces on New Survivalists.
Red chooses Settle/Consume-Trade.
Blue chooses Settle/Settle.
Red pays 3 for Galactic Bazaar.
Blue pays 2 for Imperium Warlord.
Red conquers Smuggling Lair.
Blue conquers Clandestine Uplift Lab.
Red trades good from Galactic Bazaar for 5.
Red consumes good from Smuggling Lair.
Red consumes 1 cards from hand.
Blue consumes good from New Survivalists.
Red receives 7 cards and 1 VP for Consume phase.
Blue receives 1 card and 1 VP for Consume phase.
Red chooses Develop/Settle.
Blue chooses Settle/Consume-Trade.
Red pays 5.
Red develops Pan-Galactic Research.
Blue pays 2.
Blue develops Deficit Spending.
Red pays 1 for Volcanic World.
Blue conquers Insect Uplift Race.
Red consumes good from Alpha Centauri.
Red consumes 1 cards from hand.
Blue trades good from Insect Uplift Race for 6.
Blue consumes 0 cards from hand.
Red receives 2 cards and 1 VP for Consume phase.
Blue receives 6 cards for Consume phase.
Red chooses Explore +1,+1/Produce.
Blue chooses Develop/Produce.
Red draws 5 and discards 2.
Blue draws 4 and keeps 1.
Red pays 2.
Red develops Galactic Advertisers.
Blue pays 5.
Blue develops Imperium Lords.
Red produces on Volcanic World.
Red produces on Smuggling Lair.
Blue produces on New Survivalists.
Blue produces on Insect Uplift Race.
Red receives 2 cards for Produce phase.
Blue receives 4 cards for Produce phase.
Red chooses Explore +5/Explore +1,+1.
Blue chooses Develop/Consume-Trade.
Red draws 10 and discards 7.
Blue draws 4 and keeps 1.
Red pays 5.
Red develops Prospecting Guild.
Blue pays 3.
Blue develops Drop Ships.
Red claims Expansion Leader goal.
Blue claims Expansion Leader goal.
Blue claims Greatest Military goal.
Red consumes good from Volcanic World.
Red consumes good from Smuggling Lair.
Red consumes 1 cards from hand.
Blue trades good from Insect Uplift Race for 6.
Blue consumes good from New Survivalists.
Blue consumes 2 cards from hand.
Red receives 4 cards and 2 VP for Consume phase.
Blue receives 7 cards and 3 VP for Consume phase.
Red chooses Explore +5/Develop.
Blue chooses Settle/Settle.
Red draws 9 and discards 7.
Blue draws 4 and keeps 1.
Red pays 4.
Red develops Galactic Federation.
Blue pays 6.
Blue develops Galactic Imperium.
Red pays 1 for Deserted Alien World.
Blue conquers Avian Uplift Race.
Blue claims Uplift Knowledge goal.
Blue conquers Rebel Homeworld.
Red chooses Explore +1,+1/Develop.
Blue chooses Settle/Produce.
Red draws 6 and discards 3.
Blue draws 4 and keeps 1.
Red develops Mining Conglomerate.
Blue pays 1.
Blue develops Imperium Cloaking Technology.
Blue conquers Lost Alien Warship.
Red produces on Volcanic World.
Blue produces on New Survivalists.
Blue produces on Insect Uplift Race.
Red receives 4 cards for Produce phase.
Blue receives 7 cards for Produce phase.
Blue claims Galactic Riches goal.
Blue wins with 68.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
B C Z
United States
Reston
Virginia
flag msg tools
designer
badge
Avatar
mbmbmbmbmb
Beowuulf1 wrote:
So I recently downloaded the AI. Fantastic implementation, by the way. I've been playing 2 player advanced, and am winning about 90% of the time. I'd like to be able to offer some input on the AI, but it would help if I could get a fuller log after the game that showed the cards each player had in their hand at any given time. It would also be helpful if the discarded homeworld and the initial goals were revealed as well. Is this possible?

Here is a log of a recent game. I don't know what my opponent had, but slowing to a crawl with the explore calls seemed irrational from my perspective.


Presumably you cannot see your opponent's cards because that would encourage cheating (peeking at the 'expanded' log during the game).

Perhaps an internal representation of the game could be dumped at game end (with a popup when the game ends such as 'do you want to dump the complete log' (YES) (NO) )

Still, 'helping' the AI won't work on Keldon's program, because it is self taught using learning algorithms. The reason the AI is searching is because if it sees itself in a given position and desires cards in hand it doesn't have, it will start to hunt for those cards because it has learned that that action helps it win.



1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
United States
SoCal
flag msg tools
badge
Avatar
mbmbmbmbmb
Beowuulf1 wrote:
So I recently downloaded the AI. Fantastic implementation, by the way. I've been playing 2 player advanced, and am winning about 90% of the time. I'd like to be able to offer some input on the AI, but it would help if I could get a fuller log after the game that showed the cards each player had in their hand at any given time. It would also be helpful if the discarded homeworld and the initial goals were revealed as well. Is this possible?

Here is a log of a recent game. I don't know what my opponent had, but slowing to a crawl with the explore calls seemed irrational from my perspective:




Although it's not as quick and convenient as the way you're probably suggesting, you can look this info at anytime by going to menu option Debug --> Debug Cards
and then going through that list.

Hand means in player's hand, and note the color of the player for the other field
Active means on tableau
Discard is just that
Good is just that as well, although it doesn't tell you which card is being used as the good

For discarded homeworlds, you can also look this up in the same place. Again, it's more painstaking to go through the whole list and look for homeworlds with "discarded" for its value, but it can be done
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
United States
SoCal
flag msg tools
badge
Avatar
mbmbmbmbmb
byronczimmer wrote:
Beowuulf1 wrote:
So I recently downloaded the AI. Fantastic implementation, by the way. I've been playing 2 player advanced, and am winning about 90% of the time. I'd like to be able to offer some input on the AI, but it would help if I could get a fuller log after the game that showed the cards each player had in their hand at any given time. It would also be helpful if the discarded homeworld and the initial goals were revealed as well. Is this possible?

Here is a log of a recent game. I don't know what my opponent had, but slowing to a crawl with the explore calls seemed irrational from my perspective.


Presumably you cannot see your opponent's cards because that would encourage cheating (peeking at the 'expanded' log during the game).

Perhaps an internal representation of the game could be dumped at game end (with a popup when the game ends such as 'do you want to dump the complete log' (YES) (NO) )

Still, 'helping' the AI won't work on Keldon's program, because it is self taught using learning algorithms. The reason the AI is searching is because if it sees itself in a given position and desires cards in hand it doesn't have, it will start to hunt for those cards because it has learned that that action helps it win.
Well, it's not a ladder game, or for rankings... moreso for personal use, so I don't see the harm in providing this info. If people want to brag about winning 100% of the time while using the unauthorized info, let them.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
J R

New Jersey
msg tools
mbmbmbmb
byronczimmer wrote:
Beowuulf1 wrote:
So I recently downloaded the AI. Fantastic implementation, by the way. I've been playing 2 player advanced, and am winning about 90% of the time. I'd like to be able to offer some input on the AI, but it would help if I could get a fuller log after the game that showed the cards each player had in their hand at any given time. It would also be helpful if the discarded homeworld and the initial goals were revealed as well. Is this possible?

Here is a log of a recent game. I don't know what my opponent had, but slowing to a crawl with the explore calls seemed irrational from my perspective.


Presumably you cannot see your opponent's cards because that would encourage cheating (peeking at the 'expanded' log during the game).

Perhaps an internal representation of the game could be dumped at game end (with a popup when the game ends such as 'do you want to dump the complete log' (YES) (NO) )

Still, 'helping' the AI won't work on Keldon's program, because it is self taught using learning algorithms. The reason the AI is searching is because if it sees itself in a given position and desires cards in hand it doesn't have, it will start to hunt for those cards because it has learned that that action helps it win.



If I am a better than average player, will the learning program learn quicker? I am not very knowledgable on how this type of AI works.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Serge Levert
Canada
Vancouver
British Columbia
flag msg tools
badge
Avatar
mbmbmbmbmb
Beowuulf1 wrote:
If I am a better than average player, will the learning program learn quicker? I am not very knowledgable on how this type of AI works.

I'm quite certain the AI does not model nor adapt to specific opponents. So if you sit down at your computer, or next time your friend does, the AI will treat you both the same, always. Keldon trains the AIs against each other by having them face each other millions(?) of times, then releases a version of the software. So it's not learning in real-time.
1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Randall Bart
United States
Winnetka
California
flag msg tools
designer
Baseball been bery bery good to me
badge
This is a picture of a published game designer
Avatar
mbmbmbmbmb
If the AI is only trained against mirrors of itself, it may fall into some severe group think. It's better to have two (or more) neural nets, each learning against itself and each other.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
JW
Australia
Doncaster
Victoria
flag msg tools
Avatar
mbmbmbmbmb
entranced wrote:
Keldon trains the AIs against each other by having them face each other millions(?) of times, then releases a version of the software. So it's not learning in real-time.

Barticus88 wrote:
If the AI is only trained against mirrors of itself, it may fall into some severe group think. It's better to have two (or more) neural nets, each learning against itself and each other.

This sounds like something from Terminator: The Sarah Connor Chronicles. Can the RftG AI be a potential Skynet?
1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
United States
SoCal
flag msg tools
badge
Avatar
mbmbmbmbmb
crushedguava wrote:
entranced wrote:
Keldon trains the AIs against each other by having them face each other millions(?) of times, then releases a version of the software. So it's not learning in real-time.

Barticus88 wrote:
If the AI is only trained against mirrors of itself, it may fall into some severe group think. It's better to have two (or more) neural nets, each learning against itself and each other.

This sounds like something from Terminator: The Sarah Connor Chronicles. Can the RftG AI be a potential Skynet?
If it is it won't get far. Things have a way of getting cancelled
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Front Page | Welcome | Contact | Privacy Policy | Terms of Service | Advertise | Support BGG | Feeds RSS
Geekdo, BoardGameGeek, the Geekdo logo, and the BoardGameGeek logo are trademarks of BoardGameGeek, LLC.